view mercurial/pathencode.c @ 30451:41a8106789ca

util: implement zstd compression engine

Now that zstd is vendored and being built (in some configurations), we
can implement a compression engine for zstd!

The zstd engine is a little different from existing engines. Because
it may not always be present, we have to defer load the module in case
importing it fails. We facilitate this via a cached property that holds
a reference to the module or None. The "available" method is
implemented to reflect reality.

The zstd engine declares its ability to handle bundles using the
"zstd" human name and the "ZS" internal name. The latter was chosen
because internal names are 2 characters (by only convention I think)
and "ZS" seems reasonable.

The engine, like others, supports specifying the compression level.
However, there are no consumers of this API that yet pass in that
argument. I have plans to change that, so stay tuned.

Since all we need to do to support bundle generation with a new
compression engine is implement and register the compression engine,
bundle generation with zstd "just works!" Tests demonstrating this
have been added.

How does performance of zstd for bundle generation compare? On the
mozilla-unified repo, `hg bundle --all -t <engine>-v2` yields the
following on my i7-6700K on Linux:

engine        CPU time     bundle size   vs orig size   throughput
none            97.0s     4,054,405,584     100.0%       41.8 MB/s
bzip2 (l=9)    393.6s       975,343,098      24.0%       10.3 MB/s
gzip (l=6)     184.0s     1,140,533,074      28.1%       22.0 MB/s
zstd (l=1)     108.2s     1,119,434,718      27.6%       37.5 MB/s
zstd (l=2)     111.3s     1,078,328,002      26.6%       36.4 MB/s
zstd (l=3)     113.7s     1,011,823,727      25.0%       35.7 MB/s
zstd (l=4)     116.0s     1,008,965,888      24.9%       35.0 MB/s
zstd (l=5)     121.0s       977,203,148      24.1%       33.5 MB/s
zstd (l=6)     131.7s       927,360,198      22.9%       30.8 MB/s
zstd (l=7)     139.0s       912,808,505      22.5%       29.2 MB/s
zstd (l=12)    198.1s       854,527,714      21.1%       20.5 MB/s
zstd (l=18)    681.6s       789,750,690      19.5%        5.9 MB/s

On compression, zstd for bundle generation delivers:

* better compression than gzip with significantly less CPU utilization
* better than bzip2 compression ratios while still being significantly
  faster than gzip
* ability to aggressively tune compression level to achieve
  significantly smaller bundles

That last point is important. With clone bundles, a server can
pre-generate a bundle file, upload it to a static file server, and
redirect clients to transparently download it during clone. The server
could choose to produce a zstd bundle with the highest compression
settings possible. This would take a very long time - a magnitude
longer than a typical zstd bundle generation - but the result would
be hundreds of megabytes smaller! For the clone volume we do at
Mozilla, this could translate to petabytes of bandwidth savings
per year and faster clones (due to smaller transfer size).

I don't have detailed numbers to report on decompression. However,
zstd decompression is fast: >1 GB/s output throughput on this machine,
even through the Python bindings. And it can do that regardless of the
compression level of the input. By the time you have enough data to
worry about overhead of decompression, you have plenty of other things
to worry about performance wise.

zstd is wins all around. I can't wait to implement support for it
on the wire protocol and in revlogs.

/*
 pathencode.c - efficient path name encoding

 Copyright 2012 Facebook

 This software may be used and distributed according to the terms of
 the GNU General Public License, incorporated herein by reference.
*/

/*
 * An implementation of the name encoding scheme used by the fncache
 * store.  The common case is of a path < 120 bytes long, which is
 * handled either in a single pass with no allocations or two passes
 * with a single allocation.  For longer paths, multiple passes are
 * required.
 */

#define PY_SSIZE_T_CLEAN
#include <Python.h>
#include <assert.h>
#include <ctype.h>
#include <stdlib.h>
#include <string.h>

#include "util.h"

/* state machine for the fast path */
enum path_state {
	START,   /* first byte of a path component */
	A,       /* "AUX" */
	AU,
	THIRD,   /* third of a 3-byte sequence, e.g. "AUX", "NUL" */
	C,       /* "CON" or "COMn" */
	CO,
	COMLPT,  /* "COM" or "LPT" */
	COMLPTn,
	L,
	LP,
	N,
	NU,
	P,       /* "PRN" */
	PR,
	LDOT,    /* leading '.' */
	DOT,     /* '.' in a non-leading position */
	H,       /* ".h" */
	HGDI,    /* ".hg", ".d", or ".i" */
	SPACE,
	DEFAULT  /* byte of a path component after the first */
};

/* state machine for dir-encoding */
enum dir_state {
	DDOT,
	DH,
	DHGDI,
	DDEFAULT
};

static inline int inset(const uint32_t bitset[], char c)
{
	return bitset[((uint8_t)c) >> 5] & (1 << (((uint8_t)c) & 31));
}

static inline void charcopy(char *dest, Py_ssize_t *destlen, size_t destsize,
                            char c)
{
	if (dest) {
		assert(*destlen < destsize);
		dest[*destlen] = c;
	}
	(*destlen)++;
}

static inline void memcopy(char *dest, Py_ssize_t *destlen, size_t destsize,
                           const void *src, Py_ssize_t len)
{
	if (dest) {
		assert(*destlen + len < destsize);
		memcpy((void *)&dest[*destlen], src, len);
	}
	*destlen += len;
}

static inline void hexencode(char *dest, Py_ssize_t *destlen, size_t destsize,
			     uint8_t c)
{
	static const char hexdigit[] = "0123456789abcdef";

	charcopy(dest, destlen, destsize, hexdigit[c >> 4]);
	charcopy(dest, destlen, destsize, hexdigit[c & 15]);
}

/* 3-byte escape: tilde followed by two hex digits */
static inline void escape3(char *dest, Py_ssize_t *destlen, size_t destsize,
			   char c)
{
	charcopy(dest, destlen, destsize, '~');
	hexencode(dest, destlen, destsize, c);
}

static Py_ssize_t _encodedir(char *dest, size_t destsize,
                             const char *src, Py_ssize_t len)
{
	enum dir_state state = DDEFAULT;
	Py_ssize_t i = 0, destlen = 0;

	while (i < len) {
		switch (state) {
		case DDOT:
			switch (src[i]) {
			case 'd':
			case 'i':
				state = DHGDI;
				charcopy(dest, &destlen, destsize, src[i++]);
				break;
			case 'h':
				state = DH;
				charcopy(dest, &destlen, destsize, src[i++]);
				break;
			default:
				state = DDEFAULT;
				break;
			}
			break;
		case DH:
			if (src[i] == 'g') {
				state = DHGDI;
				charcopy(dest, &destlen, destsize, src[i++]);
			}
			else state = DDEFAULT;
			break;
		case DHGDI:
			if (src[i] == '/') {
				memcopy(dest, &destlen, destsize, ".hg", 3);
				charcopy(dest, &destlen, destsize, src[i++]);
			}
			state = DDEFAULT;
			break;
		case DDEFAULT:
			if (src[i] == '.')
				state = DDOT;
			charcopy(dest, &destlen, destsize, src[i++]);
			break;
		}
	}

	return destlen;
}

PyObject *encodedir(PyObject *self, PyObject *args)
{
	Py_ssize_t len, newlen;
	PyObject *pathobj, *newobj;
	char *path;

	if (!PyArg_ParseTuple(args, "O:encodedir", &pathobj))
		return NULL;

	if (PyBytes_AsStringAndSize(pathobj, &path, &len) == -1) {
		PyErr_SetString(PyExc_TypeError, "expected a string");
		return NULL;
	}

	newlen = len ? _encodedir(NULL, 0, path, len + 1) : 1;

	if (newlen == len + 1) {
		Py_INCREF(pathobj);
		return pathobj;
	}

	newobj = PyBytes_FromStringAndSize(NULL, newlen);

	if (newobj) {
		assert(PyBytes_Check(newobj));
		Py_SIZE(newobj)--;
		_encodedir(PyBytes_AS_STRING(newobj), newlen, path,
			   len + 1);
	}

	return newobj;
}

static Py_ssize_t _encode(const uint32_t twobytes[8], const uint32_t onebyte[8],
			  char *dest, Py_ssize_t destlen, size_t destsize,
			  const char *src, Py_ssize_t len,
			  int encodedir)
{
	enum path_state state = START;
	Py_ssize_t i = 0;

	/*
	 * Python strings end with a zero byte, which we use as a
	 * terminal token as they are not valid inside path names.
	 */

	while (i < len) {
		switch (state) {
		case START:
			switch (src[i]) {
			case '/':
				charcopy(dest, &destlen, destsize, src[i++]);
				break;
			case '.':
				state = LDOT;
				escape3(dest, &destlen, destsize, src[i++]);
				break;
			case ' ':
				state = DEFAULT;
				escape3(dest, &destlen, destsize, src[i++]);
				break;
			case 'a':
				state = A;
				charcopy(dest, &destlen, destsize, src[i++]);
				break;
			case 'c':
				state = C;
				charcopy(dest, &destlen, destsize, src[i++]);
				break;
			case 'l':
				state = L;
				charcopy(dest, &destlen, destsize, src[i++]);
				break;
			case 'n':
				state = N;
				charcopy(dest, &destlen, destsize, src[i++]);
				break;
			case 'p':
				state = P;
				charcopy(dest, &destlen, destsize, src[i++]);
				break;
			default:
				state = DEFAULT;
				break;
			}
			break;
		case A:
			if (src[i] == 'u') {
				state = AU;
				charcopy(dest, &destlen, destsize, src[i++]);
			}
			else state = DEFAULT;
			break;
		case AU:
			if (src[i] == 'x') {
				state = THIRD;
				i++;
			}
			else state = DEFAULT;
			break;
		case THIRD:
			state = DEFAULT;
			switch (src[i]) {
			case '.':
			case '/':
			case '\0':
				escape3(dest, &destlen, destsize, src[i - 1]);
				break;
			default:
				i--;
				break;
			}
			break;
		case C:
			if (src[i] == 'o') {
				state = CO;
				charcopy(dest, &destlen, destsize, src[i++]);
			}
			else state = DEFAULT;
			break;
		case CO:
			if (src[i] == 'm') {
				state = COMLPT;
				i++;
			}
			else if (src[i] == 'n') {
				state = THIRD;
				i++;
			}
			else state = DEFAULT;
			break;
		case COMLPT:
			switch (src[i]) {
			case '1': case '2': case '3': case '4': case '5':
			case '6': case '7': case '8': case '9':
				state = COMLPTn;
				i++;
				break;
			default:
				state = DEFAULT;
				charcopy(dest, &destlen, destsize, src[i - 1]);
				break;
			}
			break;
		case COMLPTn:
			state = DEFAULT;
			switch (src[i]) {
			case '.':
			case '/':
			case '\0':
				escape3(dest, &destlen, destsize, src[i - 2]);
				charcopy(dest, &destlen, destsize, src[i - 1]);
				break;
			default:
				memcopy(dest, &destlen, destsize,
					&src[i - 2], 2);
				break;
			}
			break;
		case L:
			if (src[i] == 'p') {
				state = LP;
				charcopy(dest, &destlen, destsize, src[i++]);
			}
			else state = DEFAULT;
			break;
		case LP:
			if (src[i] == 't') {
				state = COMLPT;
				i++;
			}
			else state = DEFAULT;
			break;
		case N:
			if (src[i] == 'u') {
				state = NU;
				charcopy(dest, &destlen, destsize, src[i++]);
			}
			else state = DEFAULT;
			break;
		case NU:
			if (src[i] == 'l') {
				state = THIRD;
				i++;
			}
			else state = DEFAULT;
			break;
		case P:
			if (src[i] == 'r') {
				state = PR;
				charcopy(dest, &destlen, destsize, src[i++]);
			}
			else state = DEFAULT;
			break;
		case PR:
			if (src[i] == 'n') {
				state = THIRD;
				i++;
			}
			else state = DEFAULT;
			break;
		case LDOT:
			switch (src[i]) {
			case 'd':
			case 'i':
				state = HGDI;
				charcopy(dest, &destlen, destsize, src[i++]);
				break;
			case 'h':
				state = H;
				charcopy(dest, &destlen, destsize, src[i++]);
				break;
			default:
				state = DEFAULT;
				break;
			}
			break;
		case DOT:
			switch (src[i]) {
			case '/':
			case '\0':
				state = START;
				memcopy(dest, &destlen, destsize, "~2e", 3);
				charcopy(dest, &destlen, destsize, src[i++]);
				break;
			case 'd':
			case 'i':
				state = HGDI;
				charcopy(dest, &destlen, destsize, '.');
				charcopy(dest, &destlen, destsize, src[i++]);
				break;
			case 'h':
				state = H;
				memcopy(dest, &destlen, destsize, ".h", 2);
				i++;
				break;
			default:
				state = DEFAULT;
				charcopy(dest, &destlen, destsize, '.');
				break;
			}
			break;
		case H:
			if (src[i] == 'g') {
				state = HGDI;
				charcopy(dest, &destlen, destsize, src[i++]);
			}
			else state = DEFAULT;
			break;
		case HGDI:
			if (src[i] == '/') {
				state = START;
				if (encodedir)
					memcopy(dest, &destlen, destsize, ".hg",
						3);
				charcopy(dest, &destlen, destsize, src[i++]);
			}
			else state = DEFAULT;
			break;
		case SPACE:
			switch (src[i]) {
			case '/':
			case '\0':
				state = START;
				memcopy(dest, &destlen, destsize, "~20", 3);
				charcopy(dest, &destlen, destsize, src[i++]);
				break;
			default:
				state = DEFAULT;
				charcopy(dest, &destlen, destsize, ' ');
				break;
			}
			break;
		case DEFAULT:
			while (inset(onebyte, src[i])) {
				charcopy(dest, &destlen, destsize, src[i++]);
				if (i == len)
					goto done;
			}
			switch (src[i]) {
			case '.':
				state = DOT;
				i++;
				break;
			case ' ':
				state = SPACE;
				i++;
				break;
			case '/':
				state = START;
				charcopy(dest, &destlen, destsize, '/');
				i++;
				break;
			default:
				if (inset(onebyte, src[i])) {
					do {
						charcopy(dest, &destlen,
							 destsize, src[i++]);
					} while (i < len &&
						 inset(onebyte, src[i]));
				}
				else if (inset(twobytes, src[i])) {
					char c = src[i++];
					charcopy(dest, &destlen, destsize, '_');
					charcopy(dest, &destlen, destsize,
						 c == '_' ? '_' : c + 32);
				}
				else
					escape3(dest, &destlen, destsize,
						src[i++]);
				break;
			}
			break;
		}
	}
done:
	return destlen;
}

static Py_ssize_t basicencode(char *dest, size_t destsize,
			      const char *src, Py_ssize_t len)
{
	static const uint32_t twobytes[8] = { 0, 0, 0x87fffffe };

	static const uint32_t onebyte[8] = {
		1, 0x2bff3bfa, 0x68000001, 0x2fffffff,
	};

	Py_ssize_t destlen = 0;

	return _encode(twobytes, onebyte, dest, destlen, destsize,
		       src, len, 1);
}

static const Py_ssize_t maxstorepathlen = 120;

static Py_ssize_t _lowerencode(char *dest, size_t destsize,
			       const char *src, Py_ssize_t len)
{
	static const uint32_t onebyte[8] = {
		1, 0x2bfffbfb, 0xe8000001, 0x2fffffff
	};

	static const uint32_t lower[8] = { 0, 0, 0x7fffffe };

	Py_ssize_t i, destlen = 0;

	for (i = 0; i < len; i++) {
		if (inset(onebyte, src[i]))
			charcopy(dest, &destlen, destsize, src[i]);
		else if (inset(lower, src[i]))
			charcopy(dest, &destlen, destsize, src[i] + 32);
		else
			escape3(dest, &destlen, destsize, src[i]);
	}

	return destlen;
}

PyObject *lowerencode(PyObject *self, PyObject *args)
{
	char *path;
	Py_ssize_t len, newlen;
	PyObject *ret;

	if (!PyArg_ParseTuple(args, "s#:lowerencode", &path, &len))
		return NULL;

	newlen = _lowerencode(NULL, 0, path, len);
	ret = PyBytes_FromStringAndSize(NULL, newlen);
	if (ret)
		_lowerencode(PyBytes_AS_STRING(ret), newlen, path, len);

	return ret;
}

/* See store.py:_auxencode for a description. */
static Py_ssize_t auxencode(char *dest, size_t destsize,
			    const char *src, Py_ssize_t len)
{
	static const uint32_t twobytes[8];

	static const uint32_t onebyte[8] = {
		~0U, 0xffff3ffe, ~0U, ~0U, ~0U, ~0U, ~0U, ~0U,
	};

	return _encode(twobytes, onebyte, dest, 0, destsize, src, len, 0);
}

static PyObject *hashmangle(const char *src, Py_ssize_t len, const char sha[20])
{
	static const Py_ssize_t dirprefixlen = 8;
	static const Py_ssize_t maxshortdirslen = 68;
	char *dest;
	PyObject *ret;

	Py_ssize_t i, d, p, lastslash = len - 1, lastdot = -1;
	Py_ssize_t destsize, destlen = 0, slop, used;

	while (lastslash >= 0 && src[lastslash] != '/') {
		if (src[lastslash] == '.' && lastdot == -1)
			lastdot = lastslash;
		lastslash--;
	}

#if 0
	/* All paths should end in a suffix of ".i" or ".d".
           Unfortunately, the file names in test-hybridencode.py
           violate this rule.  */
	if (lastdot != len - 3) {
		PyErr_SetString(PyExc_ValueError,
				"suffix missing or wrong length");
		return NULL;
	}
#endif

	/* If src contains a suffix, we will append it to the end of
	   the new string, so make room. */
	destsize = 120;
	if (lastdot >= 0)
		destsize += len - lastdot - 1;

	ret = PyBytes_FromStringAndSize(NULL, destsize);
	if (ret == NULL)
		return NULL;

	dest = PyBytes_AS_STRING(ret);
	memcopy(dest, &destlen, destsize, "dh/", 3);

	/* Copy up to dirprefixlen bytes of each path component, up to
	   a limit of maxshortdirslen bytes. */
	for (i = d = p = 0; i < lastslash; i++, p++) {
		if (src[i] == '/') {
			char d = dest[destlen - 1];
			/* After truncation, a directory name may end
			   in a space or dot, which are unportable. */
			if (d == '.' || d == ' ')
				dest[destlen - 1] = '_';
			/* The + 3 is to account for "dh/" in the beginning */
			if (destlen > maxshortdirslen + 3)
				break;
			charcopy(dest, &destlen, destsize, src[i]);
			p = -1;
		}
		else if (p < dirprefixlen)
			charcopy(dest, &destlen, destsize, src[i]);
	}

	/* Rewind to just before the last slash copied. */
	if (destlen > maxshortdirslen + 3)
		do {
			destlen--;
		} while (destlen > 0 && dest[destlen] != '/');

	if (destlen > 3) {
		if (lastslash > 0) {
			char d = dest[destlen - 1];
			/* The last directory component may be
			   truncated, so make it safe. */
			if (d == '.' || d == ' ')
				dest[destlen - 1] = '_';
		}

		charcopy(dest, &destlen, destsize, '/');
	}

	/* Add a prefix of the original file's name. Its length
	   depends on the number of bytes left after accounting for
	   hash and suffix. */
	used = destlen + 40;
	if (lastdot >= 0)
		used += len - lastdot - 1;
	slop = maxstorepathlen - used;
	if (slop > 0) {
		Py_ssize_t basenamelen =
			lastslash >= 0 ? len - lastslash - 2 : len - 1;

		if (basenamelen > slop)
			basenamelen = slop;
		if (basenamelen > 0)
			memcopy(dest, &destlen, destsize, &src[lastslash + 1],
				basenamelen);
	}

	/* Add hash and suffix. */
	for (i = 0; i < 20; i++)
		hexencode(dest, &destlen, destsize, sha[i]);

	if (lastdot >= 0)
		memcopy(dest, &destlen, destsize, &src[lastdot],
			len - lastdot - 1);

	assert(PyBytes_Check(ret));
	Py_SIZE(ret) = destlen;

	return ret;
}

/*
 * Avoiding a trip through Python would improve performance by 50%,
 * but we don't encounter enough long names to be worth the code.
 */
static int sha1hash(char hash[20], const char *str, Py_ssize_t len)
{
	static PyObject *shafunc;
	PyObject *shaobj, *hashobj;

	if (shafunc == NULL) {
		PyObject *hashlib, *name = PyBytes_FromString("hashlib");

		if (name == NULL)
			return -1;

		hashlib = PyImport_Import(name);
		Py_DECREF(name);

		if (hashlib == NULL) {
			PyErr_SetString(PyExc_ImportError, "hashlib");
			return -1;
		}
		shafunc = PyObject_GetAttrString(hashlib, "sha1");
		Py_DECREF(hashlib);

		if (shafunc == NULL) {
			PyErr_SetString(PyExc_AttributeError,
					"module 'hashlib' has no "
					"attribute 'sha1'");
			return -1;
		}
	}

	shaobj = PyObject_CallFunction(shafunc, "s#", str, len);

	if (shaobj == NULL)
		return -1;

	hashobj = PyObject_CallMethod(shaobj, "digest", "");
	Py_DECREF(shaobj);
	if (hashobj == NULL)
		return -1;

	if (!PyBytes_Check(hashobj) || PyBytes_GET_SIZE(hashobj) != 20) {
		PyErr_SetString(PyExc_TypeError,
				"result of digest is not a 20-byte hash");
		Py_DECREF(hashobj);
		return -1;
	}

	memcpy(hash, PyBytes_AS_STRING(hashobj), 20);
	Py_DECREF(hashobj);
	return 0;
}

#define MAXENCODE 4096 * 4

static PyObject *hashencode(const char *src, Py_ssize_t len)
{
	char dired[MAXENCODE];
	char lowered[MAXENCODE];
	char auxed[MAXENCODE];
	Py_ssize_t dirlen, lowerlen, auxlen, baselen;
	char sha[20];

	baselen = (len - 5) * 3;
	if (baselen >= MAXENCODE) {
		PyErr_SetString(PyExc_ValueError, "string too long");
		return NULL;
	}

	dirlen = _encodedir(dired, baselen, src, len);
	if (sha1hash(sha, dired, dirlen - 1) == -1)
		return NULL;
	lowerlen = _lowerencode(lowered, baselen, dired + 5, dirlen - 5);
	auxlen = auxencode(auxed, baselen, lowered, lowerlen);
	return hashmangle(auxed, auxlen, sha);
}

PyObject *pathencode(PyObject *self, PyObject *args)
{
	Py_ssize_t len, newlen;
	PyObject *pathobj, *newobj;
	char *path;

	if (!PyArg_ParseTuple(args, "O:pathencode", &pathobj))
		return NULL;

	if (PyBytes_AsStringAndSize(pathobj, &path, &len) == -1) {
		PyErr_SetString(PyExc_TypeError, "expected a string");
		return NULL;
	}

	if (len > maxstorepathlen)
		newlen = maxstorepathlen + 2;
	else
		newlen = len ? basicencode(NULL, 0, path, len + 1) : 1;

	if (newlen <= maxstorepathlen + 1) {
		if (newlen == len + 1) {
			Py_INCREF(pathobj);
			return pathobj;
		}

		newobj = PyBytes_FromStringAndSize(NULL, newlen);

		if (newobj) {
			assert(PyBytes_Check(newobj));
			Py_SIZE(newobj)--;
			basicencode(PyBytes_AS_STRING(newobj), newlen, path,
				    len + 1);
		}
	}
	else
		newobj = hashencode(path, len + 1);

	return newobj;
}